Featured image source: Alexa Steinbrück / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
CHRIS LOVEDAY, VICE PRINCIPAL (BUSINESS SERVICES), BARTON PEVERIL SIXTH FORM COLLEGE, UK
Introduction
This case study builds directly on Barton Peveril Sixth Form College’s (BPSFC) initial journey with artificial intelligence (AI) and the development of bespoke AI agents, as outlined in the earlier Impact special issue case study (Loveday, 2025). While the earlier work focused primarily on operational efficiency, ethical safeguards and early pilot activity, this paper examines the subsequent scaling phase: the transition from controlled pilots to widespread student access.
Specifically, it explores how Barton Buddy and Barton AI evolved from limited trials into more widely used digital learning tools, with Barton Buddy now supporting approximately 2,200 distinct, regular student users. Following a programme of AI ‘literacy training’ delivered to all 5,000 students at the college, we have subsequently rolled out free student access to Barton AI. The case study focuses on the pedagogical implications of this growth, including student agency, creativity, digital equity and the deliberate design of a bespoke large language model (LLM) front end with enhanced guardrails and safety features.
Context and continuity
By mid-2024, BPSFC had established a secure, GDPR (General Data Protection Regulation)-compliant AI ecosystem, including Barton AI (a bespoke front end to a large language model – like ChatGPT or Gemini) and Barton Buddy (a student-facing digital assistant). Initial access was intentionally restricted to red-teamed pilots, involving Extended Project Qualification (EPQ) students and selected tutor groups of both first- and second-year students. The task was to test the agents and to probe the outputs that can be enlisted to ensure that safeguarding, accuracy and misuse risks were identified and addressed before wider roll-out.
The college’s next strategic challenge was not technical but pedagogical: how to scale access without undermining learning, creativity or student independence. Senior leaders recognised that unrestricted access to commercial AI tools, often unmoderated, opaque and inequitable, posed risks to both educational quality and student wellbeing. In response, BPSFC adopted a train-first, release-second model. This meant holding back the release of Barton AI by five months, despite it being ready, because we felt that the priority was to have delivered training sessions that better prepared students for access to a college-endorsed LLM.
Whole-college AI literacy as a pedagogical intervention
Rather than framing AI as a productivity shortcut, BPSFC positioned AI literacy as a core learning entitlement. During the first term of the 2025/26 academic year, all 5,000 students received structured training on AI and the safe, ethical and effective use of LLMs.
The training programme focused on:
- what generative AI is (and is not)
- how LLMs generate responses
- limitations, hallucinations and bias
- appropriate academic use versus misuse
- creativity as co-creation, and not substitution
- critical evaluation of AI-generated outputs.
Importantly, this training was delivered before universal student access to Barton AI was enabled. This sequencing was deemed crucial. Early student feedback demonstrated that despite clear communication about Barton Buddy’s capabilities, many learners assumed that Barton Buddy itself, as ‘AI’, had unrestricted internet access, a misconception also observed in earlier pilots. During the delivery of the training, explicit instruction helped students to reconceptualise AI as a tool requiring judgement, rather than an authority to which to be deferred.
From a pedagogical perspective, the training aimed to preserve student agency, ensuring that AI use supported ideation, explanation and exploration, rather than replacing thinking or creativity.
Rolling out free and equitable access to Barton AI
Following completion of the training programme, BPSFC made the strategic decision to provide free access to Barton AI for all students. This decision was grounded in concerns around digital equity. Without institutional provision, students with paid subscriptions to commercial AI tools would gain an unfair advantage, potentially widening attainment gaps.
Barton AI’s architecture allowed the college to avoid per-user licensing costs, relying instead on application programming interface (API)-based token usage, simply connecting an existing system or software through a direct connection. This approach proved significantly more cost-effective at scale. More importantly, the bespoke front end enabled the college to embed educational guardrails that aligned with its values and safeguarding responsibilities.
Key features included:
- secure sign-on
- no image or video generation
- no responses to questions around pre-identified ‘inappropriate subjects’, with inappropriate questions being immediately flagged by email to the safeguarding and leadership team
- the college’s filtering and monitoring systems continuing to run in the background of the agent
- full logging of prompts and outputs for accountability
- accessible design, with light and dark mode, voice to text, multi-lingual and scalable front end as a URL (meaning that it can be used on any device comfortably)
- operation entirely within the college’s secure digital tenancy, with no data being shared or training the models.
These design decisions reduced misuse risks while reinforcing the message that Barton AI was an educational environment, and not a general-purpose LLM.
Organic growth and student adoption
Following full roll-out, Barton Buddy and Barton AI experienced organic, student-led growth rather than mandated usage. By the end of the academic year (2024/25), system analytics showed approximately 2,700 distinct, regular student users accessing Barton Buddy weekly or more frequently, while use of Barton AI has continued to grow week on week.
From reviewing the usage analytics and anecdotal conversations, this adoption pattern appears significant for two reasons. First, it suggests that students perceived genuine value, particularly in:
- clarifying concepts
- supporting independent study
- navigating college systems
- drafting and refining ideas rather than final answers.
Second, the absence of compulsion preserved learner autonomy. Students chose when and how to use the relevant AI tools, reinforcing its role as a supportive scaffold rather than a required intermediary.
Qualitative feedback gathered through student discussion and a presentation from the student leadership team indicated that students increasingly used Barton AI as a ‘thinking partner’, asking for alternative explanations, testing understanding or generating prompts for further research.
Creativity, agency and pedagogical risk
According to Urmeneta and Romero (2025), one of the central tensions in AI adoption is the perceived threat to creativity. BPSFC addressed this explicitly through policy, training and system design. Students were consistently encouraged to treat AI outputs as starting points, and not finished products.
Staff guidance reinforced that:
- Originality lies in selection, synthesis and critique
- AI-generated content must be transformed, contextualised and referenced
- Creative thinking includes questioning AI responses, and not accepting them.
By embedding these expectations institutionally, the college framed AI as a creative amplifier, particularly for students who previously struggled to articulate ideas or overcome ‘blank page’ anxiety.
At the same time, risks were acknowledged. Overreliance, reduced resilience and shortcut behaviours remain live concerns. However, the combination of controlled access, explicit training and transparent system design mitigated these risks more effectively than reliance on external, commercial tools beyond institutional oversight.
Supporting wellbeing and inclusion
An unanticipated benefit of Barton Buddy’s growth was its role in low-level wellbeing support. While not designed as a counselling tool, Barton Buddy provided consistent signposting to internal and external resources and support services, particularly outside normal college hours. For some students, the ability to ask questions anonymously reduced barriers to seeking help.
It is important to emphasise that Barton Buddy is not intended to replace professional wellbeing or safeguarding support, but rather to act as a supplementary tool for signposting and initial guidance, with clear pathways to human intervention where needed.
As previously mentioned, accessibility was also enhanced, through multilingual support, voice-to-text functionality and customisable display modes. These features particularly benefit students with SEND (special educational needs and disabilities) and those using English as an additional language, reinforcing the inclusive potential of bespoke AI systems when thoughtfully designed.
Reflections and implications for practice
This case study highlights several implications for institutions considering large-scale student AI access:
- Training must precede access: AI literacy is not optional if student agency and creativity are to be preserved
- Bespoke systems enable pedagogical alignment: Front-end design choices materially shape how students engage with AI
- Equity requires institutional provision: Free access reduces the risk of AI becoming another axis of disadvantage
- Organic adoption is preferable to mandate: Voluntary use supports autonomy and meaningful engagement
- Creativity can be protected through culture, not prohibition: Clear expectations and reflective practice matter more than bans.
Implications for teachers: Introducing AI in the classroom
For teachers, introducing AI is most effective when approached as a staged, pedagogically led process rather than a technical roll-out. As seen at BPSFC, AI literacy should precede classroom use, ensuring that students understand how AI generates responses, as well as its limitations, including bias and hallucination. Early activities that involve evaluating and critiquing AI outputs can help to establish this foundation.
In practice, AI should be framed as a thinking partner rather than an answer generator. Teachers can model its use to support idea generation, alternative explanations and structuring responses, while reinforcing that final work must reflect students’ own thinking, judgement and synthesis. Designing tasks that require students to transform, critique and build upon AI outputs, rather than reproduce them, is key to maintaining academic integrity and developing higher-order skills.
Maintaining student agency remains central. This involves encouraging students to question AI outputs, ensuring that AI use does not remove productive struggle and positioning AI as an optional tool rather than a required intermediary. Alongside this, AI can play a valuable role in supporting accessibility. Features such as voice to text, rephrasing and multilingual support can help to reduce barriers for students with SEND or those using English as an additional language, provided that they are used to enhance understanding rather than replace it.
Ultimately, the impact of AI in the classroom depends less on the technology itself and more on how it is framed and used. Clear expectations, consistent modelling and a focus on student ownership of learning are essential to ensuring that AI enhances – rather than diminishes – creativity, independence and inclusion.
Conclusion
The expansion of Barton Buddy and Barton AI from pilot tools to widely adopted student platforms represents another phase in BPSFC’s AI journey. By prioritising training, safeguarding, equity and pedagogy, the college demonstrated that it is possible to scale AI use without sacrificing creativity or agency.
As AI becomes increasingly embedded in education, this case study suggests that the question is not whether students will use AI, but whether institutions will shape that use intentionally, ethically and creatively. BPSFC’s experience offers one possible model for doing so.
The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for Education or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety standards in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk.
